Improved channel quality indicator estimation using extended Kalman filter in LTE networks under diverse mobility models

Accurate channel quality indicator (CQI) estimation is crucial for optimizing resource allocation, improving link adaptation, and sustaining high performance in long term evolution (LTE) networks. In real-world scenarios, where channel conditions fluctuate rapidly due to user mobility, inaccurate CQI estimation can lead to suboptimal scheduling, degraded throughput, and reduced quality of service (QoS) for both users and network operators. Traditional Kalman filter (KF) approaches often struggle with the non-linear and time-varying nature of wireless channels, especially under unpredictable mobility patterns. This paper proposes an improved CQI estimation method based on the extended Kalman filter (EKF), which models non-linear system dynamics more effectively. The method is implemented in LTE-Sim, analyzed using MATLAB, and evaluated under random and Manhattan mobility models. Results show that across mobility regimes, KF outperforms EKF in the structured Manhattan model, while in the non-linear random-direction model, EKF yields markedly higher signal-to-interference-plus-noise ratio (SINR) stability and robustness to channel variation with SINR values above 10 dB between 300-450 s and a peak of approximately 60 dB. These results underscore the importance of mobility-aware estimation strategies in enhancing LTE network adaptability and throughput.

1. TELKOMNIKA Telecommunication Computing Electronics and Control
Vol. 23, No. 5, October 2025, pp. 1166~1176
ISSN: 1693-6930, DOI: 10.12928/TELKOMNIKA.v23i5.27205  1166

Journal homepage: http://journal.uad.ac.id/index.php/TELKOMNIKA
Improved channel quality indicator estimation using extended
Kalman filter in LTE networks under diverse mobility models


Hilary U. Ezea
1
, Mamilus A. Ahaneku
2
, Vincent C. Chijindu
2
, Obinna M. Ezeja
2
, Udora N. Nwawelu
2

1
Department of Electrical and Electronic Engineering, Faculty of Engineering, Federal University, Oye-Ekiti, Nigeria
2
Department of Electronic and Computer Engineering, Faculty of Engineering, University of Nigeria, Enugu, Nigeria


Article Info ABSTRACT
Article history:
Received Mar 9, 2025
Revised Aug 28, 2025
Accepted Sep 10, 2025

Accurate channel quality indicator (CQI) estimation is crucial for optimizing
resource allocation, improving link adaptation, and sustaining high
performance in long term evolution (LTE) networks. In real-world scenarios,
where channel conditions fluctuate rapidly due to user mobility, inaccurate
CQI estimation can lead to suboptimal scheduling, degraded throughput, and
reduced quality of service (QoS) for both users and network operators.
Traditional Kalman filter (KF) approaches often struggle with the non-linear
and time-varying nature of wireless channels, especially under unpredictable
mobility patterns. This paper proposes an improved CQI estimation method
based on the extended Kalman filter (EKF), which models non-linear system
dynamics more effectively. The method is implemented in LTE-Sim,
analyzed using MATLAB, and evaluated under random and Manhattan
mobility models. Results show that across mobility regimes, KF outperforms
EKF in the structured Manhattan model, while in the non-linear random-
direction model, EKF yields markedly higher signal-to-interference-plus-
noise ratio (SINR) stability and robustness to channel variation with SINR
values above 10 dB between 300-450 s and a peak of approximately 60 dB.
These results underscore the importance of mobility-aware estimation
strategies in enhancing LTE network adaptability and throughput.
Keywords:
Channel quality indicator
estimation
Extended Kalman filter
Long term evolution
Mobility models
Signal-to-interference-plus-
noise ratio
This is an open access article under the CC BY-SA license.

Corresponding Author:
Hilary U. Ezea
Department of Electrical and Electronic Engineering, Faculty of Engineering, Federal University Oye Ekiti
Ekiti State, Nigeria
Email: [email protected]


1. INTRODUCTION
In long term evolution (LTE) networks and other wireless mobile communication networks,
accurate knowledge of channel conditions is fundamental to ensuring efficient spectrum utilization, optimal
scheduling, and consistent service quality [1]. In modeling wireless communication systems, understanding
the inherent characteristics of the channel helps in capturing the dynamism associated with channel state
variation. The channel characteristics could be quantified as channel quality indicator (CQI) or channel state
information (CSI). While the CSI provides detailed information about the channel conditions and is used for
link adaptation, beamforming, and other transmission techniques, the CQI provides a quantized measure of
channel quality and is used to determine the modulation and coding scheme (MCS) most suitable for signal
transmission [2]. In LTE systems, CQI estimation plays a pivotal role in dynamic resource allocation,
enabling the eNodeB to assign appropriate MCS based on real-time channel conditions. Accurate CQI
reporting is crucial for maintaining quality of service (QoS), reducing retransmissions, and ensuring spectral
efficiency, particularly in heterogeneous networks and for users located in regions with poor signal coverage,
such as cell edges. In other words, CQI accuracy directly impacts throughput and reliability [3]. Estimating

2. TELKOMNIKA Telecommun Comput El Control 

Improved channel quality indicator estimation using extended Kalman filter in … (Hilary U. Ezea)
1167
the CSI accurately in a wireless fast-fading channel is highly challenging due to its complexity and the
associated level of uncertainties, unlike the CQI, which is relatively straightforward. However, CQI
estimation in mobile environments poses a significant challenge due to the highly dynamic nature of wireless
channels, which are influenced by factors such as fading, interference, and varying user mobility patterns [4].
These factors are more pronounced in high mobility scenarios, as it becomes more challenging to estimate
the channels accurately [5]. As users move, channel conditions can change unpredictably, resulting in
outdated or inaccurate CQI reports that degrade link adaptation and overall network performance [5], [6].
In determining channel conditions as a function of time, it is necessary to put into perspective the
reality of value depreciation or ageing arising from the effect of the difference between the time of
measurement and the time of usage of the measured values. If substantial time elapses between the
submission of the CQI report and its use in decision-making (such as scheduling decisions), the report’s
relevance may be significantly degraded, potentially leading to reduced network spectral efficiency [7].
Consequently, it is recommended that the estimation bias is very close to zero, such that the estimated value
does not deviate much from the actual condition of the channel. The traditional Kalman filter (KF) is a
model-based iterative technique that utilizes a series of observations to obtain a more accurate estimate of the
state parameters [8]. Although the KF techniques are effective for linear systems, they often exhibit reduced
accuracy in the presence of non-linear channel variations, commonly observed in random or irregular
mobility scenarios [9].
Several techniques have been adopted in wireless networks for estimating the CQI. Rao and Naidu
[10] proposed a signal-to-noise ratio (SNR) estimation algorithm for orthogonal frequency division multiple
access (OFDMA) systems in which the orthogonal frequency division multiplexing (OFDM) training
symbols are employed in evaluating the noise variance, while second-order moments of the received symbols
are used in estimating the signal plus noise power. Simulation results demonstrate comparable performance
with theoretical analysis, complemented by its outstanding performance when benchmarked against selected
estimation methods. Similarly, to improve SNR estimation in OFDM networks, Ling [11] adopted an
approach that aligned with the network’s non-linearity features by employing the extended Kalman filtering
technique. Comparative analysis revealed that the extended Kalman filter (EKF) estimator outperforms the
least squares (LS) and minimum mean square error (MMSE) techniques. In pursuit of even lower bit error
rate (BER), Kapil et al. [12] proposed a modified extended Kalman filter (MEKF) to jointly estimate the
channel response and auto-regressive (AR) model coefficients, combining the fast convergence rate of EKF
and the correlation feature of 2D interpolation using least squares (2DILS). Although it achieved lower BER
than EKF and 2DILS, MEKF is prone to estimation errors and comes with higher computational complexity.
In another study, Tang et al. [13] proposed a KF-based channel estimation method for 2×2 and 4×4
space-time block coding multiple-input and multiple-output orthogonal frequency division multiplexing
(STBC MIMO-OFDM) systems in dynamic environments, using orthogonal space-time codewords and pilot
sequences to suppress antenna interference before applying the KF’s prediction–update process with noise
suppression. This approach achieved strong BER and normalized mean square error (NMSE) performance,
but at the cost of increased computational load due to iterative KF processing and pilot design. Kumar and
Malleswari [14] integrated the EKF with a sliced multi-modulus algorithm (SMMA) for improving OFDM-
MIMO systems, outperforming traditional multi-modulus algorithms in terms of BER and inter-symbol
interference metrics. Rajender et al. [6] provides a comprehensive review of Kalman filter-based channel
estimation capabilities across OFDM and MIMO-STBC systems, highlighting both accuracy and
computational demands. Similarly, Drakshayini and Kounte [8] classified techniques into model-based and
deep learning-based categories, noting that while KF yields highly accurate estimates through iterative
observation, it comes with substantial computational complexity.
Building on the strengths of KF approaches, several works have adapted them specifically for CQI
prediction and dynamic resource optimization. For instance, Sulthana and Nakkeeran [15] addressed the
unrealistic assumption of perfect CQI in earlier research by predicting SNR from imperfect CQI using
Kalman filtering. The predicted SNR was then used to estimate transmission rates and design priority utilities
for scheduling decisions. Teixeira and Timoteo [16], LTE resource allocation was enhanced by using a KF-
based prediction method for determining the data rate. However, parameter fine-tuning was not considered.
In a related study Biswas et al. [17], multiple linear regression was used to estimate future throughput,
followed by KF correction to mitigate prediction and measurement errors. This approach delivered timely
and accurate throughput predictions without overfitting, making it suitable for energy-constrained LTE
devices, though with limited performance in highly dynamic channels. Extending the predictive framework
to spectrum management, Timóteo et al. [18] applied the Kalman-Takens filter (KTF) for real-time 5G
spectrum allocation. By minimizing root mean square error (RMSE), the method effectively captured traffic
dynamics, optimized throughput and latency, and adapted well to high-demand scenarios. Nonetheless, its
performance depends heavily on dataset-specific parameter tuning and understanding inter-parameter
dependencies.

3.  ISSN: 1693-6930
TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 5, October 2025: 1166-1176
1168
With the rise of machine learning, deep learning-based approaches for CQI and channel estimation
have been extensively investigated, offering new opportunities for pattern extraction and long-term
prediction. A comparative analysis in Jiang and Schotten [19] showed that while recurrent neural network
(RNN)-based predictors exhibit higher computational complexity than KF-based predictors, both achieve
comparable single-step accuracy, though RNNs demonstrate superior performance in multi-step prediction.
In vehicular systems, Kim and Han [4] proposed an received signal strength indicator (RSSI)-driven long
short-term memory (LSTM)-based CQI predictor that outperformed conventional time-series models, while
Qu et al. [20] proposed a temporal-spatial collaborative framework combining binary particle swarm
optimization (BPSO), max-relevance and min-redundancy (MRMR) feature selection, deep neural network
(DNN), and attention mechanisms, yielding proactive long term evolution-railway (LTE-R) base station
maintenance, though with significant computational demands. For unmanned aerial vehicle (UAV) ultra-
reliable low-latency communications, Bartoli and Marabissi [21] applied deep recurrent neural networks
(DRNNs) with LSTM, which reduces decode error probability and improves throughput. However, this
approach is limited to temporal CQI data and neglects spatial/frequency correlations.
Furthermore, Cwalina et al. [22] modeled a non-linear relationship between channel parameters and
block error rate (BLER), achieving a gain of up to 40% over linear models with low computational
complexity. Similarly, Diouf et al. [23] applied DNN and LSTM to real 4G datasets, achieving low RMSE
and strong prediction accuracy, but requiring large, high-quality datasets for training. Advanced wireless
scenarios, such as vehicle-to-vehicle (V2V), industrial IoT (IIoT), RIS-based systems, and mmWave MIMO,
have motivated the development of specialized and hybrid schemes. In V2V and IIoT networks, Liao et al.
[24] designed two Bayesian filter-based channel estimation techniques-basis extended model-unscented
Kalman filter (BEM-UKF), offering strong robustness at high complexity, and Basis Extended Model-
extended Kalman filter (BEM-EKF), with moderate robustness at lower complexity. For industrial
subnetworks, Gautam et al. [25] introduced a variational deep state space model (vDSSM) with sparse
student-t process regression and modified unscented KF, ensuring ultra-reliable BLER control despite the
need for real-time validation. In 5G/6G CSI prediction, Soszka [26] highlighted the potential of LSTM RNNs
across sub-6 GHz and mmWave, optimizing features and hidden layers but stressing the need for more
measurement-driven studies. In mmWave MIMO systems, Huang et al. [27] combined least square
estimation (LSE) and sparse message passing (SMP) to exploit channel sparsity, reaching near-cramer-rao
lower bound accuracy within five iterations, though adjacent-entry correlation remains unaddressed.
Reconfigurable intelligent surfaces (RIS)-assisted systems were targeted in Wei et al. [28], which proposed
parallel factor analysis (PARAFAC) decomposition using alternating least squares (ALS) and vector
approximate message passing (VAMP) algorithms, both of which outperformed benchmark schemes and
achieved near-perfect sum rate performance. Constraints on RIS element numbers and training symbol
lengths, as well as estimation ambiguity, were noted as limitations. Finally, Serunin et al. [29] developed a
CSI-RS-based CQI evaluation method involving noise estimation, SNR transformation, and MCS selection,
achieving accurate CQI reporting under additive white gaussian noise (AWGN) conditions, but requiring
further evaluation in complex fading environments.
The KF’s limitation, exemplified by reduced accuracy in modeling non-linear channel behavior,
motivates exploring the EKF, which uses first-order linearization to accommodate non-linear system
dynamics and has been shown to improve estimation in time-varying channels. This study proposes an EKF-
based CQI estimation approach for LTE networks and evaluates its performance against the classical KF
under two distinct mobility models: the structured Manhattan model and the unstructured random direction
model. Using LTE-Sim for signal-to-interference-plus-noise ratio (SINR) extraction and MATLAB for
analysis, the work demonstrates how the EKF technique enhances CQI estimation accuracy, particularly
under non-linear mobility conditions. The results are relevant for improving LTE network adaptability,
throughput, and QoS in real-world deployments.
The rest of the paper is structured as follows: section 2 presents the system model, providing a
detailed explanation of CQI estimation using KF and EKF; section 3 discusses the simulation results; and
section 4 concludes the paper.


2. METHODS
2.1. System model
In LTE networks, the channel quality can be a function of several time-varying factors, including
SINR, fading effects, interference, noise, and others. These factors assume a time-varying process, and as
such, channel quality is seen as a continuous variable that evolves in accordance with the dynamics defining
the wireless environment. This study models the channel estimation process using a linear state-space

4. TELKOMNIKA Telecommun Comput El Control 

Improved channel quality indicator estimation using extended Kalman filter in … (Hilary U. Ezea)
1169
framework, where the state and observation vectors describe the system’s underlying dynamics and
measurement processes, respectively [30]:

{
�
??????+1=�
??????�
??????+�
??????�
??????+�
??????�
_??????
�
??????=�
??????�
??????+�
??????�
??????+??????
_??????
(1)

Where: �
??????, �
??????, �
??????, �
??????, and �
?????? are respectively, ��,��,��,��, and �� constant matrices in
which 1≤�,�,�≤�; {�
??????} is a deterministic input sequence of �-vectors, i.e., �(�)∈ℛ
??????
; {�
_??????} is the
system noise sequence (zero-mean Gaussian white noise process); {??????
_??????} is the observation noise sequence
(zero-mean Gaussian white noise process).
The system shown in (1) could be decomposed into the state dynamics of a linear system to give a
sum of a linear deterministic system and the purely stochastic system shown in (2) and (3), respectively [30]:

{
�
??????+1=�
??????�
??????+�
??????�
??????
�
??????=�
??????�
??????+�
??????�
??????
(2)

{
�
??????+1=�
??????�
??????+�
??????�
_??????
�
??????=�
??????�
??????+??????
_??????
(3)

Therefore,

�
??????=�
??????+�
?????? (4)

�
??????=�
??????+�
?????? (5)

While �
?????? represents the system state vector which is a function of the system’s dynamics, the �
?????? is the
observation vector which is a function of measurements.

2.2. Estimation methods
To estimate the CQI effectively, two filtering approaches were employed: the KF for linear systems,
and the EKF for systems with non-linear dynamics. These filtering techniques were selected because of their
effectiveness in handling noisy measurements and their suitability for channel state estimation in wireless
communication systems. The following subsections detail their underlying principles, mathematical
formulations, and application to LTE CQI estimation.

2.2.1. Kalman filter
The KF is a computational method that uses a state-space model to estimate the state of a system or
process in the time domain. It leverages the relationship between the system’s state and measurement
equations to recursively estimate the state with minimal mean squared error [31], [32]. For dynamic systems
with inherent randomness and nonlinearity, the evolution of their state probability distribution over time can
be modeled using a set of non-linear differential equations [33]. KF are used for estimation in systems that
can be modeled with linear differential equations, where the state and measurements equations are presented
as linear functions within a state-space framework [34]. The estimation process involves a prediction-
correction cycle, guided by rules that refine the estimate [6]. The estimation accuracy of KF technique is high
due to the fact that it involves a series of measurements conducted over a period of time, which provides
statistically sufficient information for effective predictions of the current state [35].
In estimating the CQI using the KF, the prediction stage involves calculating the current CQI
perceived by the user based on the user’s CQI value from the previous transmission time interval (TTI).
Therefore, the correction stage involves minimization of the error between the observed value and the current
value [15]:

�
??????|??????−1=�
??????�
??????−1+�
?????? (6)

??????
??????=�
??????�
??????+�
?????? (7)

The (6) shows that the KF predicts the unknown state �
?????? based on preceding state at time (�−1) using the
measurement vector, ??????
??????. The �
?????? is a zero-mean system or process noise while �
?????? is the state transition matrix
[18]. Similarly, the �
?????? in (7) represents the measurement matrix, while �
?????? is a zero-mean observation noise.
The Kalman gain is meant to minimize estimation error and is given in (8) as [18]:

5.  ISSN: 1693-6930
TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 5, October 2025: 1166-1176
1170
??????
??????=�
??????|??????−1�
??????
??????
(�
??????�
??????|??????−1�
??????
??????
+�
??????)
−1
(8)

where �
??????|??????−1 is the covariance matrix at time � based on the estimate �
??????|??????−1 at time (�−1).
For the Update phase, the optimal estimate at time � is given in (9), thus:

�
??????
%
=�
??????|??????−1+??????
??????(??????
??????−�
??????�
??????|??????−1) (9)

The (10) gives the updated covariance matrix for the optimal estimate as:

�
??????=�
??????|??????−1(�−??????
??????�
??????) (10)

Where, ??????
?????? is the Kalman gain and � is the identity matrix.

2.2.2. Extended Kalman filter
The KF assumes an accurate mathematical model, but this assumption is often compromised due to
truncation errors when approximating non-linear systems with linear models [32]. Although the KF is
effective for many estimation problems, its limitation to finite-dimensional state representations makes it
unsuitable for systems with non-linear dynamics. To address this limitation, the EKF, which provides a first-
order linearization of non-linear systems, was developed. The enhancement achieved by EKF is a result of its
capability to approximate non-linear filtering problems using Taylor polynomial expansion [36]. This
linearization enables the EKF to apply the iterative and correction processes of the KF to systems that are
non-linear, such as the time-varying channels [37]. EKF comprises two stages, which include the prediction
stage and the correction stage [38].
In the prediction stage, an estimated current state of the channel and the error covariance estimate
are used to calculate the estimates for the next state [12].

�
??????|??????−1=??????(�
??????−1,�
??????,0) (11)

The (11) implies that the state transition model is a differentiable function, unlike the case of KF, where it is
defined as a linear function. Similarly, the (12) shows that the measurement model could be defined as a non-
linear function.

??????
??????=ℎ(�
??????)+�
?????? (12)

�
??????|??????−1=�
??????�
??????−1�
??????
??????
+�
??????�
??????−1�
??????
??????
(13)

The (13) provides the error covariance estimate, where �
?????? is the state transition matrix, �
??????
??????
is the transpose
of the state transition matrix and (�
??????�
??????−1�
??????
??????
) represents the covariance of the noise.
In the correction stage, the predicted estimate is subjected to a correctional process using the
observation model to minimize the error covariance of the estimator, resulting in an improved estimate, as
shown in (14). The (15) gives an updated error covariance estimate.

�
??????
%
=�
??????|??????−1+??????
??????(??????
??????−ℎ(�
??????|??????−1,0)) (14)

�
??????=�
??????|??????−1(�−??????
??????�
??????) (15)

Where ??????
?????? is the KF given by:

??????
??????=�
??????|??????−1�
??????
??????
(�
??????�
??????|??????−1�
??????
??????
+�
??????)
−1
(16)

Where �
??????�
??????|??????−1�
??????
??????
+�
?????? is the innovation covariance. The matrix inversion in (15) increases complexity, and
there is always a trade-off between computational complexity and the EKF estimation accuracy [36].

2.2.3. Determination of the key parameters in the filtering process
The performance of KF and EKF depends on the appropriate selection of parameters such as initial
state, covariance matrices, and process/measurement noise covariances. In implementing KF and EKF for

6. TELKOMNIKA Telecommun Comput El Control 

Improved channel quality indicator estimation using extended Kalman filter in … (Hilary U. Ezea)
1171
CQI estimation in this study, key parameters were chosen through a combination of empirical analysis,
simulation-based tuning, and reference to LTE specifications. The selection process is detailed as follows:
a. Initial state (�
0): for both the KF and EKF implementations, the initial state was set to the first CQI
value obtained from the LTE-Sim simulation output, with the assumption that the initial measured CQI
is a reasonable approximation of the true channel quality.
b. Initial covariance matrix (�
0): the initial error covariance matrix was chosen to be a diagonal matrix
with relatively large values, reflecting the initial uncertainty. For both estimation techniques, the same
initial covariance matrix value was used to allow for a fair comparison between KF and EKF. However,
differences in their underlying assumptions and algorithmic structures led to distinct performance
characteristics, especially in non-linear systems.
c. Process noise covariance (�): the � matrix was tuned experimentally by running LTE-Sim scenarios
with varying � values and comparing the estimated CQI against reference values. The optimal value,
minimized mean squared error (MSE), was selected to balance responsiveness to channel variations and
the smoothing of random fluctuations.
d. Measurement noise covariance (�): this was derived from the variance of CQI measurement errors in
the simulation, calculated as the variance between the simulator’s instantaneous CQI output and a
moving-average reference CQI over the same period. This derivation ensured that � accurately reflected
the inherent noise level of the CQI reporting process in the simulated LTE environment.
e. Kalman gain: in both KF and EKF, the Kalman gain was computed dynamically at each step from the
chosen �, �, and updated covariance values. No fixed gain was imposed, allowing the filter to adjust
weighting between prediction and measurement adaptively.
f. State transition and observation matrices: in KF implementation, both matrices were set to unity to
model a direct relationship between the previous and current states, as well as between the state and
observation. In contrast, in EKF implementation, the state transition Jacobian (??????
??????) and measurement
Jacobian (�
??????) were recalculated at each iteration based on the non-linear state and measurement models
derived from the channel mapping. These matrices ensured correct linearization for prediction-update
cycles.

2.3. Simulation setup
Simulations were conducted using LTE-Sim, which provided a robust platform for modeling LTE
system behavior under various scenarios. MATLAB was subsequently employed for post-simulation data
processing, statistical analysis, and visualization of the obtained results. The simulation environment was
configured to emulate realistic LTE downlink conditions.

2.3.1. Simulation parameters
The LTE-Sim simulation software was used to extract the SINR from the estimated CQI, and the
plots were carried out using MATLAB. To estimate the channel quality, the adaptive modulation and coding
(AMC) module in LTE-Sim was modified, and the estimated channel quality was used to determine the
SINR. Details of the simulation parameters are presented in Table 1.


Table 1. Simulation parameters
Parameter Value used
Cell scenario Single-cell
Cell radius 1 km
Number of RBs 50
Bandwidth 10 MHz
Frame structure FDD
UE speed 3 km/hr
Propagation model PED-A, Typical Urban
Mobility models Manhattan, random
Scheduling type Downlink scheduling algorithm with imperfect CQI (DSA)
Simulation duration 500 s


2.3.2. Mobility models
The mobility models considered in this work are the Manhattan mobility model and the random
mobility model. In the context of LTE network simulations, the Manhattan mobility and random direction
mobility models are commonly used to simulate user movement, with the Manhattan model representing
movement along a grid-like path, with temporal dependencies, geographic restrictions but with no spatial
dependencies and the random direction model representing random movement between points, with no
temporal dependency, nor spatial dependency, nor geographic restrictions [39]. Mobility models significantly

7.  ISSN: 1693-6930
TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 5, October 2025: 1166-1176
1172
affect SINR in LTE networks, since user mobility can lead to changes in signal strength as mobile user
moves away from the eNodeB, interference levels, and channel conditions such as path loss or fading [40].


3. RESULTS AND DISCUSSION
The SINR plots for the KF and the EKF estimations are shown in Figures 1 and 2, respectively. In
Figure 1, the SINR for the random mobility model drops below 10 dB for approximately 85% of the
observation period. In contrast, for the Manhattan mobility model, the SINR remains above 10 dB for over
50% of the observation period. This indicates that the KF is better at estimating the SINR for users following
the Manhattan mobility model than for those with random direction mobility. Figure 2 further illustrate that,
for users with random direction mobility, the SINR exceeds 10 dB for around 45% of the observation period,
with a peak value reaching 60 dB. Conversely, users with the Manhattan mobility model experience SINR
values dropping below 10 dB for about 70% of the observation period.




Figure 1. SINR Estimations using KF Figure 2. SINR Estimation using EKF


For the Manhattan mobility model, the SINR estimated using the EKF are higher than those
estimated with the KF at the early stages of estimation. However, this trend reverses in the later stages. This
observation may be due to non-linear start-up transients and greater initial uncertainty at the early stage,
which the EKF manages to capture more effectively. As the estimation progresses, the channel statistics
becomes more linear, making the KF more suitable for later stages. In contrast, in a random mobility model,
the movement of users are unpredictable and non-linear. In these situations, the EKF, which, is specifically
designed to handle non-linear systems, provides a more accurate estimate of the SINR. In summary, while
both KF and EKF techniques are valid options for SINR estimation, the EKF demonstrates superior
performance in LTE networks characterized by non-linear dynamics.
To contextualize the performance of the improved CQI estimation method, a comparative analysis
of the simulation results obtained was carried out using recent literature findings. Figure 1 shows that KF
estimates SINR more accurately in structured mobility patterns, aligning with the findings in [13], where KF
demonstrated strong BER and NMSE performance under structured channel conditions. However, it requires
iterative processing and pilot design. In a similar manner, [16] reported improvements in throughput and a
reduction in packet loss with LTE-Sim when using KF-based predictions, particularly in controlled or less
variable mobility scenarios, which aligns with our results from the Manhattan mobility model. On the other
hand, Figure 2 supports the findings of [11], which indicated that the EKF outperformed both LS and MMSE
methods in non-linear OFDM channels. Also, the results are consistent with [14], highlighting the EKF’s
superior ability to manage inter-symbol interference and non-linearities compared to traditional algorithms.
Our findings also align with [24], where the authors demonstrated that EKF-based estimators maintain
robustness in dynamic and unpredictable mobility environments, such as V2V and IIoT networks. In
comparison to the LSTM-based CQI predictors in [4], our approach provides a more straightforward
implementation with competitive performance, though less adaptive. Furthermore, the MEKF approach
described in [12] achieved a lower BER, but this came at the cost of significantly increased computational
overhead. In contrast, our results indicate that the KF and EKF strike a practical balance between estimation
accuracy and computational feasibility, particularly under the Manhattan and random mobility models.

8. TELKOMNIKA Telecommun Comput El Control 

Improved channel quality indicator estimation using extended Kalman filter in … (Hilary U. Ezea)
1173
In summary, our finding contributes to the existing literature by comparing KF and EKF within a
single cell LTE-Sim framework, assessed under structured and unstructured mobility model. The result
obtained confirm previous research indicating that KF remains efficient for structured mobility due to its
lower computational cost. Conversely, EKF effectively manages non-linear systems with unpredictable
mobility, providing greater stability and robustness regarding SINR in such scenarios. These comparisons
underscore that while advanced machine learning or hybrid techniques may achieve better accuracy, EKF
remains a practical and lightweight solution for real-time CQI estimation in LTE network environments with
varying mobility dynamics.


4. CONCLUSION
This study demonstrates that while both KF and EKF can effectively estimate CQI in LTE networks,
EKF offers superior robustness in scenarios with unpredictable, non-linear user mobility. Simulation results
show that the EKF achieves higher SINR stability in random direction mobility, whereas the KF is more
effective in structured mobility patterns, such as the Manhattan model. These findings suggest that EKF is
particularly beneficial for LTE networks with significant variations in user movement, as it can better adapt
to dynamic channel conditions, improving throughput and overall network performance.
For network operators, adopting mobility-aware estimation strategies such as EKF can lead to more
efficient resource allocation and enhanced quality of experience (QoE) on the part of users. Future research
will focus on optimizing EKF estimation parameters through artificial intelligence techniques and extending
the analysis to additional mobility models (Gauss-Markov mobility model and random waypoint mobility
model) to further validate its applicability in diverse wireless environments, including vehicle-to-everything
(V2X) networks and LTE-vehicular ad hoc network (VANET) hybrid networks.


FUNDING INFORMATION
This research was fully funded by the Federal Government of Nigeria through TETFUND grant for
Ph.D. studies administered by Federal University Oye Ekiti.


AUTHOR CONTRIBUTIONS STATEMENT
This journal uses the Contributor Roles Taxonomy (CRediT) to recognize individual author
contributions, reduce authorship disputes, and facilitate collaboration.

Name of Author C M So Va Fo I R D O E Vi Su P Fu
Hilary U. Ezea ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✓
Mamilus A. Ahaneku ✓ ✓ ✓ ✓ ✓ ✓ ✓
Vincent C. Chijindu ✓ ✓ ✓ ✓
Obinna M. Ezeja ✓ ✓ ✓
Udora N. Nwawelu ✓ ✓ ✓ ✓

C : Conceptualization
M : Methodology
So : Software
Va : Validation
Fo : Formal analysis
I : Investigation
R : Resources
D : Data Curation
O : Writing - Original Draft
E : Writing - Review & Editing
Vi : Visualization
Su : Supervision
P : Project administration
Fu : Funding acquisition



CONFLICT OF INTEREST STATEMENT
The authors declare that they have no known competing financial interests or personal relationships
that could have appeared to influence the work reported in this paper. Authors also state that there is no
conflict of interest.


DATA AVAILABILITY
The data that support the findings of this study are available on request from the corresponding
author, Ezea, H. U. He can be contacted via email: [email protected].

9.  ISSN: 1693-6930
TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 5, October 2025: 1166-1176
1174
REFERENCES
[1] M. I. Salman, A. M. Mansoor, H. A. Jalab, A. Q. M. Sabri, and R. Ahmed, “A joint evaluation of energy-efficient downlink
scheduling and partial CQI feedback for LTE video transmission,” Wireless Personal Communications, vol. 98, no. 1, pp. 189-
211, Jan. 2018, doi: 10.1007/s11277-017-4863-2.
[2] M. Sauter, From GSM to LTE-advanced pro and 5G: An introduction to mobile networks and mobile broadband. Wiley, 2017,
doi: 10.1002/9781119346913.
[3] C. Kurnaz, A. F. Kola, and M. O. Esenalp, “Performance analysis and modeling based on LTE-A field measurements: a city
center example,” International Journal of Information Technology (Singapore), vol. 15, no. 4, pp. 1919–1925, Apr. 2023, doi:
10.1007/s41870-023-01254-2.
[4] J. Kim and D. S. Han, “Deep learning-based channel quality indicators prediction for vehicular communication,” ICT Express,
vol. 9, no. 1, pp. 116–121, Feb. 2023, doi: 10.1016/j.icte.2022.05.002.
[5] J. Wu and P. Fan, “A survey on high mobility wireless communications: challenges, opportunities and solutions,” IEEE Access,
vol. 4, pp. 450–476, 2016, doi: 10.1109/ACCESS.2016.2518085.
[6] G. Rajender, T. Anilkumar, and K. S. Rao, “Estimation and performance - Kalman based approach: A review,” in International
Conference on Communication and Signal Processing, ICCSP 2016, IEEE, Apr. 2016, pp. 1145–1151, doi:
10.1109/ICCSP.2016.7754332.
[7] G. Pocovi, A. A. Esswie, and K. I. Pedersen, “Channel quality feedback enhancements for accurate URLLC link adaptation in 5G
systems,” in IEEE Vehicular Technology Conference, IEEE, May 2020, pp. 1 –6, doi: 10.1109/VTC2020-
Spring48590.2020.9128909.
[8] M. N. Drakshayini and M. R. Kounte, “A review of wireless channel estimation techniques: challenges and solutions,”
International Journal of Wireless and Mobile Computing, vol. 23, no. 2, p. 193, 2022, doi: 10.1504/ijwmc.2022.10051265.
[9] M. Zhang, K. Li, B. Hu, and C. Meng, “Comparison of Kalman filters for inertial integrated navigation,” Sensors (Switzerland),
vol. 19, no. 6, p. 1426, Mar. 2019, doi: 10.3390/s19061426.
[10] D. S. Rao and G. Naidu, “SNR estimation algorithm for OFDM systems,” International Journal of Emerging Technologies in
Engineering Research (IJETER), vol. 4, no. 6, pp. 65–69, 2016.
[11] S. O. A. Ling, “SNR estimation using extended Kalman filter technique for orthogonal frequency division multiplexing (OFDM)
system,” no. July. Universiti Tun Hussein Onn Malaysia, p. 24, 2012.
[12] J. Kapil, G. Sah, S. Aich, H. C. Kim, and P. M. Pradhan, “3GPP LTE downlink channel estimation in high-mobility environment
using modified extended Kalman filter,” in IEEE Region 10 Annual International Conference, Proceedings/TENCON, IEEE, Oct.
2018, pp. 1015–1020, doi: 10.1109/TENCON.2018.8650329.
[13] R. Tang, X. Zhou, and C. Wang, “Kalman filter channel estimation in 2 × 2 and 4 × 4 STBC MIMO-OFDM systems,” IEEE
Access, vol. 8, pp. 189089–189105, 2020, doi: 10.1109/ACCESS.2020.3027377.
[14] K. V. S. S. Kumar and B. L. Malleswari, “Channel estimation using extended kalman filter with sliced multi modulus blind
equalization algorithm (SMMA),” Journal of Theoretical and Applied Information Technology, vol. 90, no. 2, pp. 228–235, 2016.
[15] S. F. Sulthana and R. Nakkeeran, “Development and analysis of downlink scheduling algorithm in LTE system with imperfect
channel quality indicator,” International Journal of Advanced Intelligence Paradigms, vol. 18, no. 1, p. 79, 2021, doi:
10.1504/IJAIP.2021.112023.
[16] M. J. Teixeira and V. S. Timoteo, “Using a Kalman filter to improve schedulers performance in mobile networks,” in 2019 15th
International Wireless Communications and Mobile Computing Conference, IWCMC 2019, IEEE, Jun. 2019, pp. 853–858, doi:
10.1109/IWCMC.2019.8766610.
[17] M. Biswas, A. Chakraborty, and B. Palit, “A Kalman filter based low complexity throughput prediction algorithm for 5G cellular
networks,” IEEE Transactions on Vehicular Technology, vol. 73, no. 5, pp. 7089–7101, May 2024, doi:
10.1109/TVT.2023.3339649.
[18] V. S. Timóteo, M. Teixeira, and A. Sousa, “Time Series analysis in mobile communications systems,” INFOCOMP Journal of
Computer Science, vol. 24, no. 1, 2025.
[19] W. Jiang and H. Di. Schotten, “A comparison of wireless channel predictors: Artificial Intelligence versus Kalman filter,” in IEEE
International Conference on Communications, IEEE, May 2019, pp. 1–6, doi: 10.1109/ICC.2019.8761308.
[20] J. Qu, F. Liu, Y. Ma, and J. Fan, “Temporal-spatial collaborative prediction for LTE-R communication quality based on deep
learning,” IEEE Access, vol. 8, pp. 94817–94832, 2020, doi: 10.1109/ACCESS.2020.2995478.
[21] G. Bartoli and D. Marabissi, “CQI prediction through recurrent neural network for UAV control information exchange under
URLLC regime,” IEEE Transactions on Vehicular Technology, vol. 71, no. 5, pp. 5101–5110, May 2022, doi:
10.1109/TVT.2022.3152408.
[22] K. K. Cwalina, P. Rajchowski, A. Olejniczak, O. Błaszkiewicz, and R. Burczyk, “Channel state estimation in lte-based
heterogenous networks using deep learning,” Sensors, vol. 21, no. 22, p. 7716, Nov. 2021, doi: 10.3390/s21227716.
[23] N. Diouf, M. Ndong, D. Diop, K. Talla, M. Sarr, and A. C. Beye, “Channel quality prediction in 5G LTE small cell mobile
network using deep learning,” in 2022 9th International Conference on Soft Computing and Machine Intelligence, ISCMI 2022,
IEEE, Nov. 2022, pp. 15–20, doi: 10.1109/ISCMI56532.2022.10068487.
[24] Y. Liao, X. Shen, G. Sun, X. Dai, and S. Wan, “EKF/UKF-based channel estimation for robust and reliable communications in
V2V and IIoT,” Eurasip Journal on Wireless Communications and Networking, vol. 2019, no. 1, p. 144, 2019, doi:
10.1186/s13638-019-1424-2.
[25] P. Gautam, R. S. B. A. G, P. Baracca, C. Bockelmann, T. Wild, and A. Dekorsy, “CQI-based interference prediction for link
adaptation in industrial sub-networks,” Jul. 2025, [Online]. Available: http://arxiv.org/abs/2507.14169
[26] M. Soszka, “Fading channel prediction for 5G and 6G mobile communication systems,” International Journal of Electronics and
Telecommunications, vol. 68, no. 1, pp. 153–160, Dec. 2022, doi: 10.24425/ijet.2022.139863.
[27] C. Huang, L. Liu, C. Yuen, and S. Sun, “Iterative channel estimation using LSE and sparse message passing for mmWave MIMO
systems,” IEEE Transactions on Signal Processing, vol. 67, no. 1, pp. 245–249, Jan. 2019, doi: 10.1109/TSP.2018.2879620.
[28] L. Wei, C. Huang, G. C. Alexandropoulos, C. Yuen, Z. Zhang, and M. Debbah, “Channel estimation for RIS-empowered multi-
user MISO wireless communications,” IEEE Transactions on Communications, vol. 69, no. 6, pp. 4144–4157, Jun. 2021, doi:
10.1109/TCOMM.2021.3063236.
[29] I. Serunin, A. Pudeev, J. Y. Hwang, S. W. Lee, and A. Maltsev, “An overview and performance analysis of CQI reporting in 5G
NR sidelink,” in International Conference on ICT Convergence, IEEE, Oct. 2022, pp. 34 –39, doi:
10.1109/ICTC55196.2022.9952605.
[30] C. K. Chui and G. Chen, “Kalman filter: An elementary approach,” in Kalman Filtering: with Real-Time Applications, Springer,

10. TELKOMNIKA Telecommun Comput El Control 

Improved channel quality indicator estimation using extended Kalman filter in … (Hilary U. Ezea)
1175
2017, pp. 19–31.
[31] M. Zivkovic and R. Mathar, “Preamble-based SNR estimation in frequency selective channels for wireless OFDM systems,” in
IEEE Vehicular Technology Conference, IEEE, Apr. 2009, pp. 1–5, doi: 10.1109/VETECS.2009.5073813.
[32] H. Song and S. Hu, “Open problems in applications of the Kalman filtering algorithm,” in Proceedings of the 2019 International
Conference on Mathematics, Big Data Analysis and Simulation and Modelling (MBDASM 2019), Paris, France: Atlantis Press,
2019, doi: 10.2991/mbdasm-19.2019.43.
[33] M. S. Grewal and A. P. Andrews, Kalman filtering: Theory and practice with MATLAB®: Fourth edition, vol. 9781118851.
Wiley, 2014, doi: 10.1002/9781118984987.
[34] H. Musoff and P. Zarchan, Fundamentals of Kalman filtering: A practical approach, third edition. Reston ,VA: American
Institute of Aeronautics and Astronautics, 2009, doi: 10.2514/4.867200.
[35] S. H. Lee, I. K. Lim, and J. K. Lee, “Method for improving indoor positioning accuracy using extended Kalman filter,” Mobile
Information Systems, vol. 2016, pp. 1–15, 2016, doi: 10.1155/2016/2369103.
[36] Y. Liao, G. Sun, Z. Cai, X. Shen, and Z. Huang, “Nonlinear Kalman filter-based robust channel estimation for high mobility
OFDM systems,” IEEE Transactions on Intelligent Transportation Systems, vol. 22, no. 11, pp. 7219–7231, Nov. 2021, doi:
10.1109/TITS.2020.3004587.
[37] X. Dai, W. Zhang, J. Xu, J. E. Mitchell, and Y. Yang, “Kalman interpolation filter for channel estimation of LTE downlink in
high-mobility environments,” Eurasip Journal on Wireless Communications and Networking, vol. 2012, no. 1, p. 232, Dec. 2012,
doi: 10.1186/1687-1499-2012-232.
[38] O. A. Dourado, A. D. D. Neto, and P. J. Alsina, “State space estimator for plunger lift,” Journal of Petroleum Science and
Engineering, vol. 161, pp. 147–157, Feb. 2018, doi: 10.1016/j.petrol.2017.11.025.
[39] S. K. Ray, N. Z. Jhanjhi, and A. Hossain, “A reliable target evolved node B selection scheme in LTE-advanced handover,” arXiv
preprint, pp. 1–32, May 2024, [Online]. Available: http://arxiv.org/abs/2405.13827
[40] J. Parikh and A. Basu, “Effect of mobility on SINR in long term evolution systems,” ICTACT Journal on Communication
Technology, vol. 7, no. 1, pp. 1239–1244, Mar. 2016, doi: 10.21917/ijct.2016.0182.


BIOGRAPHIES OF AUTHORS


Hilary U. Ezea obtained his B.Eng. in Electrical and Electronic Engineering from
Nnamdi Azikiwe University, Awka, Anambra State, Nigeria and M.Eng. (Communication
Engineering) from University of Nigeria, Nsukka. He is a Ph.D. Student of Communication in
Electronic and Computer Engineering Department, University of Nigeria Nsukka. He is also
an academic staff in the Department of Electrical and Electronic Engineering, Federal
University Oye-Ekiti, Nigeria. He is a member of IEEE and IEEE Communication Society,
with research interests in modelling of wireless communication networks, security
engineering, renewable energy, WSN, IoT and V2X. He has quite a good number of
publications in reputable journals. He can be contacted at email: [email protected].


Mamilus A. Ahaneku received his B.Eng. (Electrical and Electronic
Engineering) and M.Sc. (Communications Engineering) from Federal University of
Technology Owerri, Imo State, Nigeria. He obtained his Ph.D. in Communications
Engineering from University of Nigeria, Nsukka in 2015. He is a Professor of
Communications Engineering, University of Nigeria Nsukka, Enugu State, Nigeria. He has
successfully supervised many Ph.D. and Masters Students. His research interests include:
wireless communication, renewable energy, microwaves system, radio frequency design and
internet of things (IoT) and image processing. He is a member of Nigerian Society of
Engineers and Nigerian Institute of Electrical and Electronic Engineers. He has published over
50 papers in reputable, high impact journals. He can be contacted at email:
[email protected].


Vincent C. Chijindu obtained his B.Eng. (Electrical and Electronic Engineering)
and M.Eng. (Computer Science and Engineering) from Anambra State University of
Technology Enugu and Enugu State University of Science and Technology, respectively. He
obtained his Ph.D. in Computer Engineering from Nnamdi Azikiwe University, Awka,
Anambra State, Nigeria in 2016. He is an Associate Professor of Computer Engineering,
University of Nigeria Nsukka, Enugu State, Nigeria. He has successfully supervised and
graduated 3 Ph.D. and 7 Masters Students, while currently supervising a handful of students at
both the Masters and Ph.D. levels. His research interests include: digital image processing,
machine learning and artificial intelligence, renewable energy systems and materials, wireless
sensor networks and systems. He received the Certificate of Merit award as the Best
Graduating Student in the Department of Electrical and Electronic Engineering, Anambra State
University of Technology in 1988. He is a member of Nigerian Society of Engineers, IEEE
Nigeria Section, and Nigeria Computer Society (NCS). He has over 45 journal publications
and one patented work to his credit. He can be contacted at email:
[email protected].

11.  ISSN: 1693-6930
TELKOMNIKA Telecommun Comput El Control, Vol. 23, No. 5, October 2025: 1166-1176
1176

Obinna M. Ezeja received his Doctorate degree (Doctor of Engineering) in
Communications Engineering from the Department of Electronic and Computer Engineering,
Faculty of Engineering, University of Nigeria, Nsukka. Prior to this, he obtained his Master of
Engineering (Communications Engineering), and Bachelor of Engineering in the same
Department and Institution. He is a Senior Lecturer in the same Department and has taught
both postgraduate and undergraduate courses in Communications Engineering. He has also
successfully supervised postgraduate and undergraduate research and project works, and is
currently the Coordinator of Graduate Studies in the Department. He has published many peer-
reviewed journal articles and conference papers. His current research interests include machine
learning-based WSNs optimization. He is also a registered member of COREN (Council for
the Regulation of Engineering in Nigeria). He can be contacted at email:
[email protected].


Udora N. Nwawelu received his B. Eng. in Electronic Engineering from the
University of Nigeria, Nsukka, Enugu State, in 2007. He holds a Master’s and a Ph.D. degrees
in Communication Engineering from the mentioned University in 2012 and 2019, respectively.
He is a Senior Lecturer in the Department of Electronic and Computer Engineering of the
University of Nigeria, Nsukka, where he has been since 2017. His area of specialization
includes developing models for network resource allocation and management, radio-
localization in wireless networks, and applications of the internet of things. He has published
several scholarly works in reputable journals. He can be contacted at email:
[email protected].